"Models may self-correct, but they’re optimised to validate the prompter’s input. When agreement is rewarded over challenge, true model independence is constrained."

Jessica Talisman critiques a VC-driven rebranding of established knowledge management ideas as “context graphs” without proper attribution. She argues this appropriation severs connections to foundational work (Polanyi, PKO, PERKS, W3C standards), undermines verification and correction mechanisms, and degrades the provenance in training data for AI. She urges builders to use existing ontologies, elicitation methods, and standards rather than marketing-driven abstractions.